12th Speech in Noise Workshop, 9-10 January 2020, Toulouse, FR

The effects of linguistic variability and CI processing on voice cue perception

Thomas Koelewijn(a), Floor Arts
University Medical Center Groningen, University of Groningen, Department of Otolaryngology, Netherlands

Etienne Gaudrain(b)
CNRS, Lyon Neuroscience Research Center, France | University Medical Center Groningen, University of Groningen, Department of Otolaryngology, Netherlands

Terrin N. Tamati
The Ohio State University, United States

Deniz Başkent(b)
University Medical Center Groningen, University of Groningen, Department of Otolaryngology, The Netherlands

(a) Presenting
(b) Attending

For cochlear-implant (CI) users, speech perception can be challenging, especially in adverse listening conditions (e.g., cocktail party situations). Talkers’ voices play an important role in speech perception, since identifying individual talkers can facilitate speech communication in challenging conditions. While previous research has suggested that several linguistic factors broadly influence talker perception, how these factors influence perception of the individual voice cues, and how this is affected by CI processing remains unclear.

The current study investigated the role of linguistic variability in voice cue perception, specifically fundamental frequency (F0) and vocal-tract length (VTL). For normal hearing adults, Just-Noticeable-Differences (JNDs) were obtained using a 3AFC adaptive paradigm. Effects of word status (words, nonwords), token identity (whether the same word/nonword was repeated in the three intervals [fixed], or different items were used [variable]), word characteristics (lexical frequency, neighborhood density), and nonword characteristics (phonotactic probability, neighborhood density) were examined.
Results demonstrated that voice cue perception in normal hearing participants was primarily affected by token identity. JNDs for F0 and VTL were significantly lower for fixed than for variable words and nonwords. This effect was largest for words. For word characteristics, F0 and VTL JNDs were affected by phonological information in nonwords, i.e., phonotactic probability, but not by lexical information, i.e., lexical frequency and neighborhood density. However, VTL JNDs varied less across linguistic conditions than F0 JNDs, suggesting different processing mechanisms for these voice cues. These findings show that linguistic variation interferes with the perception of voice cues, and that the perception of individual voice cues may be closely related to phonological processing.

The observed interactions may be different when listening to degraded speech. How the effect of linguistic variation on F0 and VTL voice cue perception is affected by CI processing was addressed in a follow-up experiment using a similar design, again with normal hearing participants, but this time including CI simulated speech. Preliminary results will be discussed. These outcomes will provide better insight into the interaction of voice cues and linguistic information, which may also improve our understanding of speech perception processes in populations with limitations in voice perception, such as CI users.

Last modified 2020-01-06 19:23:55